DeepSeek V4.1 Flash tech report: KV cache compression lets 552B beat 1.6T
机器之心 · wechat · 2026-09-12
Machine Heart's deep dive into the DeepSeek-V4.1-Flash tech report: a 552B-backbone + 196B Engram multimodal MoE that beats the 1.6T-parameter V4 Pro (scoring 40 on ArtificialAnalysis) by compressing KV cache to the extreme — global KV down to 890 bytes/token, 437x less than V1.
Key techniques:
- CED (Causal Encoder-Decoder): splits 40 layers in half; decoder KV entries are projected from layer-20 hidden states, so prefill only runs 20 layers — near 2x complexity reduction (building on Microsoft's YoCo but with per-layer projections).
- CSA2 sparse attention: fills all three compression dimensions (entry, sequence, layer) via statically assigned Full/Reindex/Reuse per-layer modes, plus FP4-quantized mainKV (E2M1 with per-16-channel E4M3 scales).
- SWA Bounded Replay: accepts approximation, replaying only the last 128 tokens and moving SWA KV out of persistent cache (DRAM pool, minutes-level TTL), cutting persistent KV to 1/8 of V4-Flash.
Other highlights: Single-Pass mHC halves activation memory traffic; the Engram N-gram-hash memory module ships in a real model for the first time, offloading static knowledge to host memory; the DSpark speculative decoding module is trained post-pretraining and accelerates both serving and RL rollouts. Post-training openly has "no algorithmic novelty," but introduces a 1–100 reasoning effort scalar: raising effort from 25 to 100 lifts average Pass@1 across eight reasoning benchmarks from 67.1% to 76.3% at 2.5x output tokens; API tiers max/high/low map to b=100/75/50. Context scaling from 4K to 1M adds only 1/4 decode FLOPs per token; despite verbosity (89k tokens/task) it costs just $0.27 per task.
Related event: DeepSeek V4.1 Flash Launches Open-Source, Beats GPT-5.6 at One-Third Cost(4 posts)→
More from Infra
- Clover-35B-A3B Quantized to 29.4 GB Runs at ~42 tok/s on a 64 GB Mac — viveky259 · 2026-10-31
- Liquid Agent brings fully local LFM3 agent to Linux with one-command install — JosephJacks_ · 2026-10-31
- A fully local office AI stack: MiniCPM, Qwen3, LFM2.5 and Whisper — HowDevelop · 2026-10-31
- a16z: GPU rental prices up 106% in 3 months while token prices fall 58% — a16z · 2026-10-31
- vLLM nightly adds PLE disk offload, with a telling [Qwen4Exp] tag on the PR — rmhubbert · 2026-10-31
- Solo AI studio: per-minute GPU rentals rendered a 328-frame Blender shot for ~$1 — r0ck3t23 · 2026-10-31